As is known, Intel acquired Mobileye and a share in the cartographic service HERE to promote its own interests in the automotive electronics segment. Moreover, the processor giant prepares for the announcement its first specialized processor for "autopilot" systems, which can be released this year. Previously it was assumed that this product will combine computing cores with x86-compatible architecture and programmable matrices, or some specialized functional blocks. Now representatives of Intel transparently hint that they have an arsenal of processor cores from Atom to Xeon, and the company's car platform will be scalable.
Along the way, the head of the "automotive" direction of Intel Jack Weast (Jack Weast) explained that at the stage of robocar development the typical data set will be divided into five main processing stages :
- Reception of data;
- data analysis;
- Preparation of data for systems with learning function;
- Application Development;
- simulation.
At the first stage, the data from the hard drive of the car is transferred to the server, and this is done in a very primitive way - the media is extracted from the car, and physically connects to the server. Now, when the autopilot systems are only being designed, for four to five hours of test trips the average prototype of the "robotmobile" is able to accumulate several tens of terabytes of data. Serial machines will need to store less information, and then it will be possible to use 5G wireless networks for data transmission. Prototypes retain a lot of background information, which can be useful for training the system.
Further, the information collected by the prototype is structured to facilitate its use. Modern systems of the autopilot are sharpened under allocation of abnormal situations - it helps to teach them to react to all possible events faster. Each piece of data must receive attributes that help to use it more effectively for learning the system.
Interestingly, unlike some automakers (Audi, for example), Intel believes in the need for a "central processor" in automotive autopilot systems. It, in particular, allows to create a panoramic model of the environment for assessing the situation and choosing the trajectory. The central processor should receive disparate information from cameras and sensors, as well as neighboring "smart" cars and road infrastructure objects.